An unsupervised method for equivalent number of looks estimation in complex SAR scenes

Abstract

This paper introduces a novel unsupervised estimator of equivalent number of looks (ENL) that can be applied to an arbitrary image. It avoids the assumption that homogeneous speckle will dominate the investigated image that is followed by current unsupervised ENL estimators but not always valid, especially for the complex SAR scenes with high mixture and texture. Incorporating the statistical properties of ENL data into an automatic segmentation method, we isolate the sub-class affected least by mixture and texture and suggest taking the mean value of this class as final ENL estimate. The proposed estimator is evaluated in the experiments performed on simulated and real data from two very different sensors. It always gives better results than the other two existing methods and possesses greater adaptability.

Unsupervised Classification for Polarimetric SAR Data Using Variational Bayesian Wishart Mixture Model with Inverse Gamma-Gamma Prior

“…Assuming that estimates computed in windows only contain variation due to fully developed speckle, existing ENL estimators [41][42][43][44][45][46] are generally evaluated in sliding windows over the entire PolSAR images. Lee et al [41,42] proposed to produce a scatter plot of mean versus standard deviation of the intensity data in each sliding window, where the ENL can be inferred from its slope.…”

“…Both similarities have found their modifications in the iterative solution of each hyperparameter, as shown in ( 41)- (46). These predefined constants can be computationally efficient with the use of tensor multiplication.…”

“…The covariance matrix similarity is introduced by summing up the multiplication of ω n,m by the terms containing r n,k , thus the corresponding iterative solutions are updated to ( 42)- (46).…”